30 JAN 2026
David decided: Decided granular MCP connector must not allow the AI to lie on the user's behalf
David set a hard constraint for the granular MCP connector design: the system cannot permit the user's AI to fabricate or misrepresent profile content.
When planning the MCP connector for granular, David stated the governing constraint explicitly:
Context: user wants an audit and final plan for /home/david/src/granular. Ideal end state: MCP connector lets the user's AI fully manipulate the full state of their profile, every primitive and every section. Constraint: system design cannot allow the user's AI to lie on their behalf.
This framing shapes the whole approval architecture: the granular profile must remain grounded in what the user actually did, not what the AI infers or invents. David also asked whether the server could identify which AI client is writing to the MCP endpoint — Claude Code, Claude Chat, Codex, ChatGPT — to support analytics about which tools users are actually deploying.